Anthropic scales Claude Mythos to critical infrastructure in 15+ countries
Anthropic is expanding Project Glasswing to about 150 organizations across more than 15 countries to use Claude Mythos for finding software vulnerabilities.
Intelligence analysis by GPT-5.4 Mini
Anthropic is widening access to its security-focused Claude Mythos model after an initial pilot with 50 partners, including the U.S. government. The new cohort spans critical sectors like power, water, healthcare, communications, and hardware across more than 15 countries.
Anthropic is giving a very smart computer helper to more groups that run important services like power, water, and hospitals so they can find hidden holes in their software, like a careful security guard checking doors and windows before a storm.
Analysis
What changed
Anthropic says it is expanding Project Glasswing, its initiative for finding and fixing critical software vulnerabilities with AI, to about 150 more organizations across more than 15 countries. The company says Claude Mythos sits at the center of the effort and describes it as its most powerful model yet for spotting vulnerabilities at scale.
Who gets access
The initial rollout in early April gave 50 partners, including the U.S. government, access to Claude Mythos Preview for scanning codebases for weaknesses. The new group is broader and includes organizations in sectors Anthropic says were not well represented before, such as power, water, healthcare, communications, and hardware. The company says many of these organizations maintain codebases that other companies and governments depend on.
Why Anthropic is pushing this now
Anthropic says the stakes are unusually high because a successful attack on these codebases could be catastrophic. In the company’s framing, a major attack could affect more than 100 million people for most partners, with consequences for global and national security. Anthropic also says it expects other AI companies to reach similar model capability soon, which is part of why it is trying to build safeguards now through Project Glasswing.
Competitive context
The expansion comes one day after Anthropic said it filed confidentially for an IPO, following a $65 billion funding round at a nearly $1 trillion valuation. The company is also facing growing competition in AI security: TechCrunch notes that OpenAI has released GPT-5.5-Cyber and is testing it with a large partner group.
The article does not fully verify every organization named in the expanded cohort, but it reports that Financial Times cited access for Okta, Samsung, SK Hynix, SK Telecom, NATO, and ENISA.
Key points
- Anthropic is expanding Project Glasswing to about 150 new organizations in more than 15 countries.
- Claude Mythos is the model at the center of the initiative and is being used for vulnerability scanning.
- The expanded group includes power, water, healthcare, communications, and hardware organizations.
- Anthropic says a major attack on many of these codebases could affect more than 100 million people.
- The move comes as Anthropic files confidentially for an IPO and faces growing AI cybersecurity competition from OpenAI.
If Project Glasswing works as intended, critical infrastructure operators could find dangerous software flaws earlier and fix them before attackers can use them. The broader rollout could also help spread stronger security practices across sectors that many other systems depend on.
The biggest risk is that the same powerful tool could still be outpaced by real attackers if defenses lag behind model capability. There is also a concentration risk: if many essential organizations rely on the same AI system, any weakness or misuse could have broad consequences.
